Mainly, though, each b+-tree node is the same size as a disk I/O block. The disk is often the slowest part of the database, so making as few reads as possible w

Mainly, though, each b+-tree node is the same size as a disk I/O block. The disk is often the slowest part of the database, so making as few reads as possible will go a long way. If you can't work entirely in memory, that's the next best thing. That makes no sense. An external hash index is also designed to do as few I/Os per read as well (typically just a single I/O to read the page that holds the target bucket). Just like b-trees are designed differently than in-memory trees, external hash indexes are designed differently than in-memory hash tables.